Automated counting of white blood cells in thin blood smear images

Blood cell counting plays a crucial role in clinical diagnosis to evaluate the overall health condition of an individual. Traditionally, blood cells are manually counted using a hemocytometer; however, this task has been found to be time-consuming and error-prone. Recently, machine learning-based approaches have been employed to effectively automate counting tasks. In this work, the fifth version of the 'you only look once' (YOLOv5) object detection method was adopted to automatically detect and count white blood cells (WBCs) in porcine blood smear images. YOLOv5 was chosen because of its speed and accuracy. The dataset used in this study was collected specifically for this WBC counting task. Our experimental results exhibit the high speed and efficiency of YOLOv5 in detecting and counting WBCs, having obtained an accuracy of 89.25% and a mean average precision at 0.5 intersection over union threshold (mAP 0.5) of 99%.